Scholarlink

<p>This study employed physiological data from wearable sensors to build deep-learning models aimed at predicting exercise exertion levels. A total of twenty-one healthy individuals participated in 16-minute cycling sessions, during which real-time data, including ECG, pulse rate, oxygen saturation, and revolutions per minute (RPM), were captured across three different intensity levels: two minutes at low intensity with no resistance, four minutes at medium resistance, and ten minutes at high intensity with maximum resistance. Each intensity level was further divided into consecutive 2-minute segments, totaling eight segments per session. For each segment, the average values of RPE, heart rate, RPM, and oxygen saturation were calculated and used as predictive features. Participants provided ratings of perceived exertion (RPE) every minute. Each session was divided into eight 2-minute segments to analyze exercise intensity. Average values heart rate, RPM, and oxygen saturation were computed for each segment to be used as predictive features. Heart rate variability (HRV) parameters were also derived from the ECG recordings. Averaged RPE served as the response variable. The Minimum Redundancy Maximum Relevance (MRMR) technique was applied to identify key predictors for both classification and regression tasks. These selected features were then used to train and validate Long Short-Term Memory (LSTM) models. The classification model demonstrated 90% accuracy and an F1-score of 0.89, while the regression model showed a mean squared error (MSE) of 1.1 in the test set.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Automated Prediction of Exercise Intensity Using Physiological Data and Deep Learning

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年3月24日

查看全文
  • Aref Smiley,
  • Joseph Finkelstein

摘要

This study employed physiological data from wearable sensors to build deep-learning models aimed at predicting exercise exertion levels. A total of twenty-one healthy individuals participated in 16-minute cycling sessions, during which real-time data, including ECG, pulse rate, oxygen saturation, and revolutions per minute (RPM), were captured across three different intensity levels: two minutes at low intensity with no resistance, four minutes at medium resistance, and ten minutes at high intensity with maximum resistance. Each intensity level was further divided into consecutive 2-minute segments, totaling eight segments per session. For each segment, the average values of RPE, heart rate, RPM, and oxygen saturation were calculated and used as predictive features. Participants provided ratings of perceived exertion (RPE) every minute. Each session was divided into eight 2-minute segments to analyze exercise intensity. Average values heart rate, RPM, and oxygen saturation were computed for each segment to be used as predictive features. Heart rate variability (HRV) parameters were also derived from the ECG recordings. Averaged RPE served as the response variable. The Minimum Redundancy Maximum Relevance (MRMR) technique was applied to identify key predictors for both classification and regression tasks. These selected features were then used to train and validate Long Short-Term Memory (LSTM) models. The classification model demonstrated 90% accuracy and an F1-score of 0.89, while the regression model showed a mean squared error (MSE) of 1.1 in the test set.